Which is better, Claude Opus 4.8 or GPT-5.2?

Across 59 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 44 and GPT-5.2 leads on 14 — Claude Opus 4.8 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Opus 4.8 vs GPT-5.2

AnthropicvsOpenAI59 shared benchmarks44–14 head-to-head

Benchmark	Claude Opus 4.8	GPT-5.2
AA Agentic Index	49.4	60.2
AA Intelligence	57.3	43.3
AA-LCR	73	79.3
AA-Omniscience	28.8	0.3
aime_2026	100	98.3
arc_agi_1	92.5	86.2
ARC-AGI-2	72.1	52.9
arena_elo	1482	1437
arena_text_factuality	1464	1442
arena_vision	1280	1245
Artificial Analysis Coding Index	74.3	48.7
browsecomp	88.5	65.8
charxiv_rq	80.5	82.1
coding_arena_elo	1539	1418
critpt	20.9	11.6
deepsearchqa_f1	93.1	71.3
Fortress	18.2	17.5
frontiermath_tier_4	31.3	18.8
gdpval	54.2	48.3
GDPval-AA (Elo)	1890	1462
GPQA Diamond	93.6	92.4
HLE	57.9	45.5
HLE (with tools)	57.9	45.5
hmmt_feb_2026	96.7	97
hmmt_nov_2025	96.5	99.2
IFBench	62.2	75.4
imo_answer_bench	83.5	86.3
livebench	77.2	74.8
livebench_agentic_coding	50.5	50.3
livebench_coding	81.8	76.1
livebench_data_analysis	66	78.2
livebench_instruction_following	72	61.8
livebench_language	79.7	79.8
livebench_math	94.3	93.2
livebench_reasoning	89.2	83.2
longbench_v2	69.1	54.5
matharena_visual_math_overall	81.6	86.5
mathvision	86.7	83
MCP Atlas	83.6	68
MMMU-Pro	78.9	79.5
MMVU	79.2	80.8
OmniScience Accuracy	48.8	44.3
OmniScience Non-Hallucination	60.7	38.4
PaperBench	80.3	63.7
scicode	53.5	52.1
screenspot_pro_no_tools	87.9	86.3
simplebench	64.8	45.8
simpleqa_verified	39.5	38.9
SWE-bench Multilingual	84.4	72
SWE-bench Pro	69.2	55.6
SWE-bench Verified	88.6	80
TauBench V3 - Banking	34.2	11.1
Terminal-Bench 2.0	74.6	54
Terminal-Bench Hard	58.3	62.2
tool_decathlon	59.9	46.3
toolathlon	59.9	46.3
Video-MME	86	86
τ²-Bench Telecom (AA run)	94.4	98.7
τ³-Bench	27.6	12

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.